编辑说明
本文梳理了塔勒布在推上关于 AI 的公开推文进行原文级分类整理 。
不做解读、不做综合、不做立场总结，仅按表层主题归类。
Category 1 — What LLMs Are
定义性陈述（LLM 是什么）
Original statement
“To make it pedagogical, LLM is an autocomplete function that randomizes based on the statistical frequency.”
中文
为了便于教学理解，大型语言模型本质上是一种基于统计频率进行随机化输出的自动补全函数。
Date
2025-07-05
Source
https://x.com/nntaleb/status/1941485474876448913
Original statement
“What we’ve learned from LLM is that when your sample is monstrously large, induction can fool you as it becomes nearly indistinguishable from deduction.”
中文
我们从大型语言模型中学到的是：当样本规模巨大时，归纳推理会欺骗你，因为它几乎与演绎推理无法区分。
Date
2025-02-28
Source
https://x.com/nntaleb/status/1895568488389099656
Original statement
“The reason LLM models feel human is that most humans rephrase combinations of things they heard/saw, with nothing new.”
中文
大型语言模型之所以让人感觉“像人”，是因为大多数人本身也只是重组、复述自己听到或看到的东西，并没有真正的新内容。
Date
2025-09-02
Source
https://x.com/nntaleb/status/1962950743104655387
Category 2 — Induction, Probability, and Scale
归纳、概率与规模效应
Original statement
“Because of contamination (LLM learning from LLM), the 85% will converge to 100%.”
中文
由于污染效应（LLM 从其他 LLM 学习），原本 85% 的概率会收敛到 100%。
Date
2025-07-05
Source
https://x.com/nntaleb/status/1941489178849517851
Original statement
“One effect of scaling: the improvement in the probability estimates… more data will reduce the variance.”
中文
规模扩展的一个结果是概率估计看起来更好……更多的数据会降低方差。
Date
2025-07-05
Source
https://x.com/nntaleb/status/1941489178849517851
Category 3 — Errors, Hallucinations, and Failures
错误、幻觉与失败案例
Original statement
“Asked ChatGPT (as well as Grok & Claude) to identify books on some shelves… it basically missed 2/3 of the books and INVENTED titles.”
中文
我让 ChatGPT（以及 Grok 和 Claude）识别书架上的书……结果它漏掉了三分之二的书，还编造了一些书名。
Date
2025-07-27
Source
https://x.com/nntaleb/status/1949517645335687665
Original statement
“Note: Google AI says mistakenly ‘The word ‘ammia’ does not appear in the Tell Amarna letters.’”
中文
注意：Google 的 AI 错误地声称“‘ammia’这个词并未出现在阿玛尔纳书信中”。
Date
2025-05-25
Source
https://x.com/nntaleb/status/1926678636955865440
Original statement
“If you want to be a scholar, do not use the ‘AI differs’ but PROVIDE EXACT QUOTATION OF SENECA.”
中文
如果你想成为一名学者，不要用“AI 给出的不同说法”，而是要提供塞涅卡的准确原文引用 。
Date
2025-10-11
Source
https://x.com/nntaleb/status/1977050596898398691
Category 4 — Permissible Uses of AI
可接受用途（Taleb 明确承认在用）
Original statement
“ChatGPT explains to me MY OWN derivations and proofs of 10 y ago… like a slightly dumb referee.”
中文
ChatGPT 向我解释了我十年前自己写过的推导和证明……就像一个有点笨的裁判。
Date
2025-09-10
Source
https://x.com/nntaleb/status/1965742640634466464
Original statement
“I used AI to explain, not to derive.”
中文
我使用 AI 是为了解释 ，而不是为了推导 。
Date
2025-04-11
Source
https://x.com/nntaleb/status/1910637064686567683
Original statement
“Hahaha! ChatGPT as a BS Buster!”
中文
哈哈！ChatGPT 作为一个“辟谣器”！
Date
2025-03-30
Source
https://x.com/nntaleb/status/1906438142065860621
Category 5 — Structural Limitations
结构性限制（不可 / 无法）
Original statement
“The highest priced skill a human can have is detecting circularity. My impression is that AI is structurally incapable of it.”
中文
人类最有价值的技能之一是识别循环论证。而我的印象是，AI 在结构上并不具备这种能力。
Date
2025-06-03
Source
https://x.com/nntaleb/status/1929974312296956183
Original statement
“If someone is telling you he understands AI, he doesn’t understand AI.”
中文
如果有人告诉你他“理解 AI”，那他其实并不理解 AI。
Date
2025-06-15
Source
https://x.com/nntaleb/status/1934319347213722089
Category 6 — Labor and Professions
劳动、职业与技能
Original statement
“AI is moving lawyers, editors, radiologists into lower value jobs… Gains are going to technofeudalists.”
中文
AI 正在把律师、编辑、放射科医生推向价值更低的工作……收益流向了技术封建主义者。
Date
2025-10-09
Source
https://x.com/nntaleb/status/1976271162943602696
Original statement
“Friends, please help complete the list of people not threatened by AI: plumber, chef, gardener…”
中文
朋友们，请帮忙补充那些不受 AI 威胁 的职业名单：水管工、厨师、园丁……
Date
2025-06-18
Source
https://x.com/nntaleb/status/1935325694990680339
Category 7 — Academia, Education, and Institutions
学术、教育与制度
Original statement
“FINALLY AI is killing academic games in finance & economics.”
中文
终于，AI 正在终结金融和经济学领域里的学术游戏。
Date
2025-12-03
Source
https://x.com/nntaleb/status/1996233031774855396
Original statement
“Even if AI cannot replace doctors (yet), it may still replace medical school… more educational than operational.”
中文
即便 AI 还不能取代医生，它也可能取代医学院……其作用更多体现在教育层面，而非操作层面。
Date
2025-08-23
Source
https://x.com/nntaleb/status/1959142692136390800
Original statement
“Molto importante! Yann LeCun is delivering the AI consequence of Quine’s Dogmas of Empiricism.”
中文
非常重要！Yann LeCun 正在阐释奎因《经验主义的两大教条》在 AI 领域的后果。
Date
2025-12-22
Source
https://x.com/nntaleb/status/2003196101646778449
Category 8 — First-Person and Contextual Remarks
第一人称与语境性评论
Original statement
“I feel intellectually safe thanks to AI; the way I feel safe cycling next to a Waymo.”
中文
由于 AI 的存在，我在智力上感到安全，就像在 Waymo 自动驾驶车辆旁骑行时感到安全一样。
Date
2025-12-15
Source
https://x.com/nntaleb/status/2000625913273229665
Original statement
“Thanks Grok… it feels relief to have AI set the record straight.”
中文
感谢 Grok……让 AI 把事实澄清，确实让人感到宽慰。
Date
2025-12-14
Source
https://x.com/nntaleb/status/2000192618404200778
Original statement
“Looks like AI is not yet fooled by the Hasbara narrative.”
中文
看起来，AI 目前还没有被 Hasbara 叙事所欺骗。
Date
2025-03-22